Google DeepMind's AlphaFold AI model, which has already revolutionized scientists' understanding of proteins, has expanded its capabilities in a new version released this week.

The big picture: The new AlphaFold 3 can predict what interactions between nearly all of the molecules that form the basis of life look like, which could open roads to new drugs or more resilient crops.

• The interactions AlphaFold 3 predicts are key for many crucial processes in cells. The interplay between and changes to proteins, DNA, RNA, ions and other small molecules dictate their function — and dysfunction from disease.
• For example, when a protein on the surface of a cell binds to another protein on a virus, the molecules change shape, setting off a process that fuses the virus and the cell so the virus can invade. The details of those interactions could help develop precise vaccines or antiviral drugs.

"Biology clearly is a dynamic system, so we need to understand interactions between different structures, proteins and other things to really understand what they do," Google DeepMind CEO Demis Hassabis told Axios.

• "AlphaFold 3 is a big step in that direction."

Driving the news: AlphaFold 3 is the next iteration of AlphaFold models that took on and solved one of biology's toughest problems: predicting the structure of proteins from their amino acid sequence.

• The new AI model handles a larger number of chemicals using a different approach.
• It leverages a generative AI technique called diffusion, which is similar to those that drive image and video generators, like DALL-E.

How it works: AlphaFold 3 takes a cloud of atoms and then refines it, step by step, until the model converges on the most accurate molecular structure it can predict.

• The number of inputs it can handle is "dramatically expanded" from earlier AlphaFold models, John Jumper, director at Google DeepMind, told Axios.
• The reported accuracy ranges from 40% to 80%, depending on the interaction AlphaFold 3 is trying to model, and the program provides a measure of how confident it is in its result. These results were published in the journal Nature.
• AlphaFold 3 performs better than existing tools for almost all categories of interactions they looked at.

"These problems we're doing, we wouldn't consider them solved," Jumper said. "We're still at an accuracy we'd like to improve."

• The tool at this point helps to more quickly — and cheaply — home in on possible structures that can be jumping-off points for more detailed studies.
• DeepMind also launched a server for researchers to access AlphaFold 3, but it has some restrictions about what can be modeled, particularly for drug candidate molecules.
• Some scientists say these restrictions could limit its impact.

Between the lines: The diffusion technique comes with a risk.

• In what's known as disordered regions, or flexible parts of a protein that can take on many shapes, the model will produce a plausible-looking structure but one that couldn't exist — a biological form of the hallucinations that plague other AI models.
• AlphaFold 3 does report its low confidence in these results, Jumper said, adding that the team has reduced that risk by adding more data to the regions where these hallucinations typically occur.

The big picture: "My dream is to build a model of a virtual cell," Hassabis told Axios, but "the challenges become almost sort of exponentially more difficult."

• "The system is going to have to learn some fundamentals about how biophysics works. We think we can do that," he said.
• "The question is getting the right amount and the right quality of data."

Experimental tools to image what's happening inside cells without killing them are being developed.

• If those tools arrive, "that will be huge for AI to then learn from that," Hassabis said. Or researchers may have to build physics simulations that can provide synthetic data.

While today's AI can already help science and medicine, as AI gets applied to solve more types of problems, the AI models themselves will improve too, Hassabis said.

The Department of Energy on Tuesday announced a sweeping artificial intelligence program that would give it a big role — and a unique one — in the federal government's AI research efforts.

Driving the news: The department announced the Frontiers in Artificial Intelligence for Science, Security and Technology (FASST) initiative at the AI Expo for National Competitiveness in Washington.

• "Imagine we had a basic science AI foundational model like ChatGPT for English — but it speaks physics and chemistry," deputy energy secretary David Turk said in announcing the initiative.
• Combine that "with the world-class laboratory test facilities we have at [DOE] labs and you will get a sense of the incredible potential here," he said, adding it is already happening with fusion ignition research at Lawrence Livermore National Lab.

Why it matters: The DOE has world-class supercomputing, a powerful scientific infrastructure and experience working with dual-use technologies that position it to power AI advances for science and national security.

• "It is arguably the most important AI initiative yet from the Biden administration" considering the ambition, scale, funding and focus squarely on AI, says Divyansh Kaushik, a VP at Beacon Global Strategies who focuses on critical and emerging tech.

Zoom in: FASST focuses on how the DOE can leverage its supercomputing resources, data it collects from scientific work at its facilities and the know-how of scientists at the 17 national labs to "supercharge a new AI initiative that will build on this infrastructure," Helena Fu, DOE's director of the Office of Critical and Emerging Technologies, tells Axios.

• The initiative is "going to be focused on solving mission-critical challenges that the private sector won't be investing in or can't invest in based on the specific interests that the U.S. government has," she said. Those include DOE's national security missions and the use of AI for scientific discovery.

Read the entire story

A new report calls on all levels of government to strengthen U.S. biodefense measures and urges policymakers to codify parts of a national strategy to address an array of biological threats.

Why it matters: Threats in the form of infectious disease outbreaks, lab accidents and biology-based weapons are expected to increase in the coming years, according to the report's authors and other experts.

• But biodefense investments get caught in a cycle of "panic and neglect" — an intense focus for a short period, after which policymakers, funders and the public move on, the report notes.
• "Every future administration must ensure that the National Biodefense Strategy keeps pace with the rapidly evolving and increasing biological threat," the authors of the 2024 National Blueprint for Biodefense write.

The report was released Tuesday by the Bipartisan Commission on Biodefense, which was set up in 2014 with former Sen. Joe Lieberman as founding co-chair. Lieberman contributed to the report before his death this year.

"We're not putting enough emphasis on getting ahead of these biological threats," says Asha George, the commission's executive director.

• "The United States has become exceedingly good at responding to all kinds of things. But I think we're at our limit now as a nation, as a world, and that means being really good at response is no longer good enough."

Zoom in: The commission calls for Congress to designate the White House's national security adviser as leader of national biodefense efforts and to establish a deputy adviser to quarterback daily duties and responsibilities.

• We are "really calling upon the national security adviser to take charge of this. There are too many national security implications," George says.
• Biodefense responsibilities currently span the government, including 15 federal departments and nine independent agencies.

Go deeper

U.S. funders to tighten oversight of controversial "gain-of-function" research (Max Kozlov — Nature)

NASA's budget woes put ambitious space research at risk (Adam Mann — Science News)

A fundamental stage of human reproduction is shifting (Katherine Wu — The Atlantic)

An AI analysis of sperm whale communications revealed an "alphabet," a team of researchers reported this week.

The big picture: There is a longstanding debate about whether language is unique to humans or something we share with other animals.

• Some scientists are using AI tools — including generative AI — to try to understand animal communication.

How it works: Sperm whales (Physeter macrocephalus) communicate with one another using different patterns of clicks known as codas — a Morse code between whales.

• Scientists know from previous research that those codas can help identify the caller but "almost everything else about the sperm whale communication system, including its structure and information-carrying capacity, remains unknown," the team from MIT and Project CETI wrote in Nature Communications.

What they found: Analyzing nearly 9,000 codas from a sperm whale clan in the Eastern Caribbean with at least 60 members, they found the duration, rhythm and tempo of the clicks vary and they could be organized into an "alphabet."

• The whales sometimes even add extra clicks — what the scientists call ornamentation. But they aren't random — codas with ornaments are often found at the begining or end of a sequence, the scientists report.
• "Sperm whale vocalisations are more expressive and structured than previously believed," they wrote.

The intrigue: What the codas and their combinations mean is an open question.

• But the researchers argue the results suggest sperm whales can combine codas in different ways — a key feature of language.